Overview of Storage and Indexing 1 Data on

Overview of Storage and Indexing 1

Data on External Storage l Disks: l l Tapes: l l Can only read pages in sequence. Cheaper than disks; used for archival storage. File organization: l l l Can retrieve random page at fixed cost. Reading several consecutive pages is much cheaper than reading them in random order. Method of arranging a file of records on external storage. Record id (rid) is sufficient to physically locate record. Buffer manager: l Software that fetches pages from external storage to main memory buffer pool. Storage and Indexing 2

Alternative File Organizations l Heap (random order) files: l l Sorted Files: l l Records are stored in random order. Best if records must be retrieved in some order, or only a range of records is needed. Indexes: l l l Data structures that organize records on disk to optimize certain kinds of retrieval operations. Like sorted files, they speed up searches for a subset of records, based on values in certain search key fields. Updates are much faster than in sorted files. Storage and Indexing 3

Indexes l l l An index on a file speeds up selections on the search key fields for the index. Any subset of the fields of a relation can be the search key for an index on the relation. Search key is not the same as key (minimal set of fields that uniquely identify a record in a relation). An index contains a collection of data entries. A data entry with search key value k is denoted as k*. Given data entry k*, we can find record with key k in at most one disk I/O. Storage and Indexing 4

Alternatives for Data Entry l l Three alternatives for storing a data entry k*: 1. Actual data record (with key value k), or 2. <k, rid> (rid is the record id of data record with search key value k), or 3. <k, rid-list> Choice of alternative for data entries is orthogonal to the indexing technique used to locate data entries with a given key value k. Examples of indexing techniques: B+ trees, hash-based structures Typically, index contains auxiliary information that directs searches to the desired data entries Storage and Indexing 5

Alternatives for Data Entries (Cont. ) l Alternative 1 (actual data record): l l l The index structure is a file organization for data records (instead of a Heap file or sorted file). At most one index on a given collection of data records can use Alternative 1. (Otherwise, data records are duplicated, leading to redundant storage and potential inconsistency. ) If data records are very large, number of pages containing data entries is high. Implies size of auxiliary information in the index is also large, typically. Storage and Indexing 6

Alternatives for Data Entries (Cont. ) l Alternatives 2 (<k, rid>) and 3 (<k, rid-list>): l l Data entries typically much smaller than data records. So, better than Alternative 1 with large data records, especially if search keys are small. Alternative 3 more compact than Alternative 2, but leads to variable sized data entries even if search keys are of fixed length. Storage and Indexing 7

B+ Tree Indexes l l Leaf pages contain data entries, and are chained. Non-leaf pages have index entries; only used to direct searches. 20 3* 15 4* data records Storage and Indexing 27 5* 7* 50 <= k 20 <= k < 50 k < 20 5 50 10* data records 16* 61 41 19* data records 21* 26* data records 75 27* 36* data records 8

Hash-Based Indexes l l Good for equality selections. Index is a collection of buckets. l l l Hashing function h: l l l Bucket = primary page plus zero or more overflow pages. Buckets contain data entries. h(r) = bucket in which (data entry for) record r belongs. h looks at the search key fields of r. No need for “index entries” in this scheme. Storage and Indexing 9

Index Classification l Primary vs. secondary l l Unique index l l If search key contains primary key, then called primary index. Search key contains a candidate key. Clustered vs. unclustered l l If order of data records is the same as, or `close to’, order of data entries, then called clustered index. Alternative 1 implies clustered; in practice, clustered also implies Alternative 1 (since sorted files are rare). A file can be clustered on at most one search key. Cost of retrieving data records through index varies greatly based on whether index is clustered or not! Storage and Indexing 10

Clustered vs. Unclustered Index l l l Suppose that Alternative 2 is used for data entries, and that the data records are stored in a Heap file. To build clustered index, first sort the Heap file (with some free space on each page for future inserts). Overflow pages may be needed for inserts. (Thus, order of data records is `close to’, but not identical to, the sort order. ) Unclustered Clustered index entries index file index entries data records Storage and Indexing data file data records 11

Cost Model l l Parameters (typical value): l B: Number of data pages l R: Number of records per page l D: Average time to read or write disk page (15 msec) l C: Average time to process a record (100 nsec) l H: Time to apply the hash function to a record (100 nsec) l F: Fan-out of tree indexes (100) Ignoring CPU costs for simplicity (ignore C and H) Measuring number of page I/O’s ignores gains of pre-fetching a sequence of pages. Average-case analysis; based on several simplistic assumptions. Storage and Indexing 12

To Compare l l File organizations l Heap files (random order; insert at eof) l Sorted files, sorted on <age, sal> l Clustered B+ tree file, Alternative 1, search key <age, sal> l Heap file with unclustered B+ tree index on search key <age, sal> l Heap file with unclustered hash index on search key <age, sal> Operations l Scan: Fetch all records from disk l Equality search l Range selection l Insert a record l Delete a record Storage and Indexing 13

Assumptions l l l Heap Files: l Equality selection on key; exactly one match. l Files compacted after deletions. Sorted Files: Indexes: l l l 80% page occupancy => File size = 1. 25 data size l Implies file size = 1. 5 data size Tree: 67% occupancy (this is typical). Scans: l l l Alternates 2, 3: data entry size = 10% size of record Hash: No overflow buckets. Leaf levels of a tree-index are chained. Index data-entries plus actual file scanned for unclustered indexes. Range searches: l We use tree indexes to restrict the set of data records fetched, but ignore hash indexes. Storage and Indexing 14

Sorted Files l Scan: B(D + RC) l l Equality search: Dlog 2 B + Clog 2 R l l l The cost of search plus the cost of retrieving the set of records that satisfy the search Insert: Search + BD l l Assuming that the equality selection matches the sort order <age, sal>. Locating the first page using a binary search at a cost of Dlog 2 B; locating the first qualifying record of the page using a binary search at a cost of Clog 2 R Range search: D(log 2 B + # pages with matching pages) l l Examining all pages. We first find the correct position in the file, add the record, and then fetch and rewrite all subsequent pages. The cost is the search cost plus the read cost 0. 5 B(D + RC) and the write cost 0. 5 B(D + RC). Delete: Search + BD l We must search for the record, remove it, and write the modified page back. We must also write all subsequent pages to compact the free space. Storage and Indexing 15

Heap File with Unclustered Tree Index l Assumption: l l Scan: BD(R + 0. 15) l l For each qualifying data entry, we incur one I/O to fetch the corresponding data records. Insert: D(3 + log. F 0. 15 B) l l We locate the first page (Dlog. F 0. 15 B), find the first qualified data entry of the page (Clog 26. 7 R), and read the data record (D). Range search: D(log. F 0. 15 B + # pages with matching records) l l The cost of reading all data entries is 0. 15 B(D + 6. 7 RC) and the cost of fetching all records is BR(D + C). Equality search: D(1 + log. F 0. 15 B) l l Each data entry in the index is a 1/10 the size of a data record. The number of leaf pages in the index (or the size of data entries) is 1. 5(0. 1 B) = 0. 15 B. The number of data entries on a page is 0. 67(10 R) = 6. 7 R. We insert the data record (2 D + C), find the right leaf page (Dlog F 0. 15 B + Clog 26. 7 R), and write it out (D). Delete: Search + 2 D l The search step costs Dlog. F 0. 15 B + Clog 26. 7 R and the step of writing out the modified pages in the index and the data file costs 2 D. Storage and Indexing 16

Cost of Operations Scan Equality Search Range Search Insert Delete Heap BD 0. 5 BD BD 2 D Search + D Sorted BD Dlog 2 B D(log 2 B + # matching pages) Search + BD Clustered 1. 5 BD Dlog. F 1. 5 B D(log. F 1. 5 B + # matching pages) Search + D Unclustered tree index BD(R + 0. 15) D(1 + log. F 0. 15 B) D(log. F 0. 15 B + # D(3 + matching records) log. F 0. 15 B) Search + 2 D Unclustered hash index BD(R + 0. 125) 2 D BD Search + 2 D Storage and Indexing 4 D Search + D 17

Understanding the Workload l For each query in the workload: l l Which relations does it access? Which attributes are retrieved? Which attributes are involved in selection/join conditions? How selective are these conditions likely to be? For each update in the workload: l l Which attributes are involved in selection/join conditions? How selective are these conditions likely to be? The type of update (INSERT/DELETE/UPDATE), and the attributes that are affected. Storage and Indexing 18

Choice of Indexes l What indexes should we create? l l Which relations should have indexes? What field(s) should be the search key? Should we build several indexes? For each index, what kind of an index should it be? l Clustered? Hash/tree? Storage and Indexing 19

Choice of Indexes (Cont. ) l One approach: Consider the most important queries in turn. Consider the best plan using the current indexes, and see if a better plan is possible with an additional index. If so, create it. l l Obviously, this implies that we must understand how a DBMS evaluates queries and creates query evaluation plans! Before creating an index, must also consider the impact on updates in the workload! l Trade-off: Indexes can make queries go faster, updates slower. Require disk space, too. Storage and Indexing 20

Index Selection Guidelines l Attributes in WHERE clause are candidates for index keys. l l l Exact match condition suggests hash index. Range query suggests tree index. Clustering is especially useful for range queries; can also help on equality queries if there are many duplicates. Storage and Indexing 21

Index Selection Guidelines (cont. ) l Multi-attribute search keys should be considered when a WHERE clause contains several conditions. l l Order of attributes is important for range queries. Such indexes can sometimes enable index-only strategies for important queries. For index-only strategies, clustering is not important! Try to choose indexes that benefit as many queries as possible. Since only one index can be clustered per relation, choose it based on important queries that would benefit the most from clustering. Storage and Indexing 22

Examples of Clustered Indexes l B+ tree index on E. age can be used to get qualifying tuples. l l l Consider the GROUP BY query. l l l How selective is the condition? Is the index clustered? If many tuples have E. age > 10, using E. age index and sorting the retrieved tuples may be costly. Clustered E. dno index may be better! Equality queries and duplicates: l Clustering on E. hobby helps! Storage and Indexing SELECT E. dno FROM Emp E WHERE E. age>40 SELECT E. dno, COUNT (*) FROM Emp E WHERE E. age>10 GROUP BY E. dno SELECT E. dno FROM Emp E WHERE E. hobby=Stamps 23

Indexes with Composite Search Keys l Composite Search Keys: Search on a combination of fields. l Equality query: Every field value is equal to a constant value. E. g. wrt <sal, age> index: l l Range query: Some field value is not a constant. E. g. : l l age=13 and sal =75 11, 80 11 12, 10 12, 20 12 13, 75 index <sal, age> age > 20; or age=20 and sal > 10 10, 12 Data entries in index sorted by 20, 12 search key to support range 75, 13 queries. l Lexicographic order, or 80, 11 l Spatial order. index Storage and Indexing <age> <age, sal> name age sal 12 bob 12 10 13 index cal 11 80 joe 12 20 sue 13 75 data <sal> 10 20 75 80 data entries in index lexicographic order 24

Composite Search Keys l l To retrieve Emp records with age = 30 AND sal = 4000, an index on <age, sal> would be better than an index on age or an index on sal. l Choice of index key orthogonal to clustering etc. l Clustered tree index on <age, sal> or <sal, age> is best. l Clustered <age, sal> index much better than <sal, age> index! If condition is: 20 < age < 30 AND 3000 < sal < 5000: If condition is: age = 30 AND 3000 < sal < 5000: Composite indexes are larger, updated more often. Storage and Indexing 25

Index-Only Plans l A number of queries can be answered without retrieving any tuples from one or more of the relations involved if a suitable index is available. SELECT E. dno, COUNT(*) FROM Emp E GROUP BY E. dno <E. dno> SELECT AVG(E. sal) FROM Emp E WHERE E. age=25 AND E. sal BETWEEN 3000 AND 5000 <E. age, E. sal> or <E. sal, E. age> SELECT E. dno, MIN(E. sal) FROM Emp E GROUP BY E. dno Tree index <E. dno, E. sal> Tree index Storage and Indexing 26

Index-Only Plans (Cont. ) l Index-only plans are possible if the key is <dno, age> or we have a tree index with key <age, dno> l Which is better? (<age, dno>) SELECT E. dno, COUNT (*) FROM Emp E WHERE E. age=30 GROUP BY E. dno l What if we consider the second query? (<dno, age>) SELECT E. dno, COUNT (*) FROM Emp E WHERE E. age>30 GROUP BY E. dno Storage and Indexing 27

Index-Only Plans (Cont. ) l Index-only plans can also be found for queries involving more than one table. SELECT D. mgr FROM Dept D, Emp E WHERE D. dno=E. dno SELECT D. mgr, E. eid FROM Dept D, Emp E WHERE D. dno=E. dno <E. dno> Avoid retrieving tuples from one relation. <E. dno, E. eid> Storage and Indexing 28